Background - ctDNA

• cell-free DNA (cfDNA) found in plasma has nuermous uses as a biomarker.
• circulating tumour DNA (ctDNA) can be used to track tumour dynamics in real-time.
• Measuring ctDNA within plasma provides a non-invasive manner in which one can track tumour progression.

Background - CNA

• Gene amplifications and deletions contribute signifigantly to tumorigenesis.
• Copy number alterations (CNA) have traditionally been detected with copartive genomic hybridisation (CGH) microarrays.
• Shallow whole genome sequencing (sWG), defined as coverage <1x, offers a cheap altertnative to microarray detection (<£50 per sample).

Background - data set

• ctDNA samples from 30 patients with metastatic breast cancer
• Various number of ctDNA samples per patient

data$meta[type=="plasma",.N,patient][,table(N)]

## N
##  3  4  5  6  7  8  9 11 12 15 17 
##  2  7  6  1  4  5  1  1  1  1  1

• ctDNA samples taken longitutidial, with the longest period spanning 478 days
• Primary tumour samples (FFPE) for 27 patients
• Buffy coat (normal) samples for 26 patients
• Sequencing depth of ranging from 0.001x to 0.05x (single-ended)
• Response Evaluation Criteria In Solid Tumours (RECIST) values based on MRI scans availible for 23

NGS methods for CNV detection

Pipeline

Alignment

• QC using trimgalore (<1% adapter)
• Alignment using bwa-mem
• Indexing and sorting with Samtools
• Dedupping with Picard (0.1% to 0.2% dups)
• Merge replicates using Samtools

CNA calling

• Binning with QDNA-seq
• GC & Mappability correction with QDNA-seq
• Blacklisting with QDNA-seq
• Segmentation with DNAcopy
• CNA calling with …. TBC

Binning

• Bin size choice: smaller bin sizes (more bins) may increase resolution of the segmentation, however, it must be balanced by the increase in varience or “noise”
• An excess of noisy bins can make the segmentation difficult, as the noise may not be Gaussian, especially when many bins have zero reads.
• 15kb bin widths for 6 million reads (~0.1x coverage) shown to work well. - QDNA paper
  • With 0.05x coverage –> ~100 reads per bin with bin width of 30 kb
  • With 0.01x coverage –> ~100 reads per bin with bin width of 100 kb

Blacklisting & Corrections

• GC content: The proportion of G and C bases in a region and the count of fragments mapped to it are directly related.
• Mappibility Correction: A measure of how often the sequence found at the particular location will align within the whole genome
• Blacklist: Areas prone to extreme values based on the analysis of a large volumnes of data (1000G). These can be either common germline copy number variations (CNVs) or artifacts of the methodlogy.

Segmentation: (C)BS

• Let \(X_1,...,X_n\) be the log ratios of the intensities, which are indexed by the locations of the \(n\) markers being studied
• Let \(S_i=X_1+...+X_i, 1<=i<= n\), be the partial sums.

\[Zi = (\frac1i + \frac1{n − i})^{−1/2}(\frac{S_i}{i} − \frac{S_n − S_i}{n − i}).\] - Let \(Z_B = max(|Z_i|)\) for \(1<i<N\). - The null hypothesis of no change is rejected if the statistic exceeds the upper αth quantile of the null distribution of ZB

## Warning: Removed 3 rows containing missing values (geom_point).

Final Output

## Warning: Removed 1108 rows containing missing values (geom_point).

## Warning: Removed 473 rows containing missing values (geom_path).

Quality control

We need a measure of the quality of each sample. However, tradiational measures, such as the variance, are affected by CNAs. The QDNA package outputs (top right) the trimmed (0.01%) variance, however, this also appears to be affected by the presence of CNAs. Within the paper, they suggest the first order difference is a better measure of the noise.

\[Noise = median(|x_{i+1}-x_i|);~1<=i<(N-1)\]

Segmentation - Sanity Check

Before we start an indepth analysis of the segmentation, lets just confirm that there are more segments found within FFPE/FF/plasma than BC samples and that they segmentation count is independent of the noise.

\[Area = \sum^N_i(|S_i|) \forall i\]

Inferring gains/losses, cellularity and abosulute copy numbers

• Cancer cells are nearly always intermixed with an unknown fraction of normal cells (purity)
• The DNA content of ctDNA from whole chromosomal abnormalties is unknown (ploidy)
• Tumours are heterogenous and various CNA may be unique to certain subclones (subclones)

Examples

ABSOLUTE - pipeline

ABSOLUTE - Example

## Warning: Removed 111 rows containing missing values (geom_point).

## Warning: Removed 38 rows containing missing values (geom_path).

ABSOLUTE - Results

x <- merge(absolute$meta[,.(Sample.name,call.status)],data$meta,all.y = T)
table(x[,.(type,call.status)],useNA = "ifany")

##         call.status
## type     called high entropy high non-clonal <NA>
##   BC          3            0               0   24
##   FFPE       32            6               0    0
##   Frozen      4            0               0    0
##   plasma    109            3               1   89

## Warning in melt.data.table(x, id.vars = c("type", "call.status"),
## measure.vars = c("area.seg", : 'measure.vars' [area.seg, n.seg, noise]
## are not all of the same type. By order of hierarchy, the molten data value
## column will be of type 'double'. All measure variables not of type 'double'
## will be coerced to. Check DETAILS in ?melt.data.table for more on coercion.

RECIST Measurements

## Warning in melt.data.table(x[!is.na(cycle)], id.vars = c("patient",
## "cycle", : 'measure.vars' [RECIST, purity, area.seg, n.seg, noise] are not
## all of the same type. By order of hierarchy, the molten data value column
## will be of type 'double'. All measure variables not of type 'double' will
## be coerced to. Check DETAILS in ?melt.data.table for more on coercion.

## Warning: Removed 35 rows containing missing values (geom_point).

Blind test

outside link

Acknowledgments

• Caldas lab
  • Meiling Gao
  • Oscar Rueda
• CRUK Core

Session Info

sessionInfo()

## R version 3.3.2 (2016-10-31)
## Platform: x86_64-apple-darwin16.1.0 (64-bit)
## Running under: macOS Sierra 10.12.3
## 
## locale:
## [1] en_GB.UTF-8/en_GB.UTF-8/en_GB.UTF-8/C/en_GB.UTF-8/en_GB.UTF-8
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] stringr_1.2.0     data.table_1.10.4 ggthemes_3.4.0    ggplot2_2.2.1    
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_0.12.11     knitr_1.16       magrittr_1.5     munsell_0.4.3   
##  [5] colorspace_1.3-2 rlang_0.1.1      plyr_1.8.4       tools_3.3.2     
##  [9] grid_3.3.2       gtable_0.2.0     htmltools_0.3.6  yaml_2.1.14     
## [13] lazyeval_0.2.0   rprojroot_1.2    digest_0.6.12    assertthat_0.2.0
## [17] tibble_1.3.3     reshape2_1.4.2   evaluate_0.10    rmarkdown_1.6   
## [21] labeling_0.3     stringi_1.1.5    scales_0.4.1     backports_1.1.0